Hello Ben,
I want to deploy and use on ROS ( melodic or noetic) a simple deep learning network I developed with python ( on spyder environment).
There are few steps:
- In your Keras model keep that file as a ROS node and then check the form of data and publish it.
- After publishing, You can use that data as whatever you want.
Just for Example (Drone with hand signs): I have hand sign data based on the video stream(right, left, up, and down) and I have obtained a Keras model for it. Then I would suggest making the full model as a single node and then publish it. After publishing let's say I subscribed and I have a string
and based on that particular string I can decide moments and give moments to the drone via tf and it performs the action.
In place of string, you can take any message like Image, transformation, coordinate, pointcloud or even you can create your custom message.
Does this make sense? If you want in-depth understanding to write a comment. This is just an overview.
Don't think about Keras, TensorFlow, OpenCV, point cloud, or any of this the most important thing you should think about is data and how you will send it to ros or your robot